Reactor Diagnostics Rule Generation Through Statistical Pattern Recognition

Publisher Website
Google Scholar
Abstract
A pattern recognition algorithm has been developed for systematic generation of shallow knowledge for nuclear power plant transient diagnostics. The algorithm involves feature selection and pattern discovery. The selection of N best features is attained by discarding redundant and nondiscriminatory features. An entropy minimax algorithm is used to discover the patterns by searching an N-dimensional feature space, populated with transient events of the data base, to locate subspaces that discriminate among the event classes. These patterns are then represented as production rules for diagnostics. A series of approximations have been implemented in the algorithm to handle the discovery of patterns in multidimensional space. We have also developed a perturbation algorithm within the entropy minimax framework to update the patterns in an incremental fashion as new data are obtained. The Midland Nuclear Power Plant Unit 2 simulator is used to generate 144 single-failure events. Based on these events, 25 production rules are generated, representing a two-level hierarchical knowledge structure of single-failure events along the critical safety function approach. These rules represent the common characteristics of time-varying features over the diagnostic time, thereby providing diagnostic capability at any time during the transient.

This publication has 0 references indexed in Scilit: